andrewluo/rfdetr-commonforms-test

RF-DETR Fine-tuned on CommonForms

This model is an RF-DETR (small) fine-tuned on the CommonForms dataset for form field detection.

Model Details

• Model Type: RF-DETR small
• Dataset: jbarrow/CommonForms
• Classes: 3
• Epochs: 1
• Batch Size: 4 (grad_accum: 4)

Classes

[ { "id": 0, "name": "class_0", "supercategory": "form_element" }, { "id": 1, "name": "class_1", "supercategory": "form_element" }, { "id": 2, "name": "class_2", "supercategory": "form_element" } ]

Usage

import torch
from PIL import Image

# Load model
model_path = "path/to/rfdetr_model.pt"
# Note: You'll need the rfdetr library installed
from rfdetr import RFDETRSmall

model = RFDETRSmall()
model.load_state_dict(torch.load(model_path))
model.eval()

# Run inference
image = Image.open("form.jpg")
predictions = model.predict(image)
print(predictions)

Training Details

• Learning Rate: 0.0001
• Effective Batch Size: 16
• Dataset: Trained on CommonForms (form field detection)

Metrics

(Add your evaluation metrics here after running evaluation)

Citation

@misc{rfdetr-commonforms,
  author = {Your Name},
  title = {RF-DETR Fine-tuned on CommonForms},
  year = {2024},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/andrewluo/rfdetr-commonforms-test}}
}